The Storage Is The Application


Coughlin, Tom

August 14, 2013

The increasing density of electronic circuits makes it possible to pack lots of memory and processing power in a single chip package.  As a result of this increasing chip complexity more and more functions can be included with modern circuitry.  This has resulted in multi-core processing and storage space for more capable and detailed firmware.  As a consequence modern electronics and the devices made with them are “smarter” that ever before.

Many analysts, academics and futurists have talked about being able to build a complete application on a chip and we have achieved the capability to finally start doing this.  One result of the greater smarts in chips is that we could build storage devices that could have capabilities normally associated with higher-level computer systems.  By making the storage devices in our data centers (and ultimately in our consumer devices-for examples see my book “Digital Storage in Consumer Electronics, Newnes Press) smarter we could help reduce general system bottlenecks.

During Ed Doller’s keynote talk at the 2013 Flash Memory Summit he talked about a further realization of smart storage devices.  This implementation puts file system capability and the associated content metadata within with a conventional SSD.  As a consequence processing is brought to the content being processed.  This allows atomic level processing with minimal latency between the data and the processing.

By having metadata and file system capabilities built into a storage device such as an SSD, parallel database searches across data stored in multiple SSDs are possible.  Also with these types of smart storage devices, adding additional storage devices doesn’t just scale out in storage capacity and traditional performance but it also allows “scaling in” for even more efficiency as coordinated smart storage devices process parallel operations between their individual data volumes.

Micron says that they have prototyped such SSDs for the data center without any additional hardware and they double overall SQL search performance and achieve 39% lower power for a search using 8 SSDs in a server.  This lower power is a result of using less system bandwidth to move data around since the data and processing are adjacent in the SSD.  They show that as the number of SSDs doing the search increases for a given size of search records the overall speed and power savings over conventional searches increases.  Extending out to 32 SSDs in a server, Micron expects an 8X search performance increase and power savings of 84%.

By tying the storage device more intimately to the application with smarter capability and knowledge on the storage device itself we can create new models for faster data centers that generate less heat and thus save on power to keep these data centers cool.  If we then use these ideas to build this sort of self-knowledge into other types of storage devices we can make faster and more efficient computers and consumer devices which use less power and thus reduce battery power.   By making the storage device the means to an application we can satisfy many important needs.

Great ! Thanks for your subscription !

You will soon receive the first Content Loop Newsletter